Gut resistome presents a unique biome signature in chronic multi-symptom illness patients and links persistent pro-inammatory phenotype in a mouse model reversible by fecal microbiota transfer

: 21 Chronic multi-symptom illness (CMI) affects a subsection of elderly and war veterans and 22 is associated with systemic inflammation, chronic fatigue, pain and neuroinflammation. 23 We showed previously that an altered gut microbiome-inflammation axis aids to the 24 symptom reporting and persistence. Here, a mouse model of CMI and a group of Gulf 25 War veterans’ with CMI showed the presence of an altered host resistome, a signature of 26 antibiotic resistance genes within the microbiome. Results showed that antibiotic 27 resistance genes were significantly altered in the CMI group in both mice and GW 28 veterans when compared to the control. Fecal samples from GW veterans with persistent 29 CMI showed a significant increase of resistance to a wide class of antibiotics and 30 exhibited an array of mobile genetic elements distinct than normal healthy controls. 31 Strikingly, the altered resistome and gene signature were correlated with mouse serum 32 IL6 levels. Altered resistome in mice also correlated strongly with intestinal inflammation, 33 decreased synaptic plasticity that was reversible with fecal microbiota transplant (FMT), 34 a tool to restore a healthy biome. The results indicate an emerging linkage of the gut 35 resistome and CMI and might be significant in understanding the risks to treating hospital 36 acquired infections in this population. 37 38 39 40 41

a large number of antibiotics [16], [17], [18]. Veterans have been reported to be resistant 66 to amoxicillin, β-lactams, fluroquinolones, methicillin and most importantly the 67 carbapenem group of antibiotics [19], [20]. In our present study, we aimed to analyze the 68 ARG and mobile genetic element (MGE) patterns in chronic multisymptom illness (CMI) 69 in a GW mouse model as well as in a cohort of GW veterans . GW illness (GWI) is a linked to chemical exposures experienced during the war. [21,22], [23]. 76 Several studies in preclinical mouse models of CMI related to GW have reported that 77 exposure to chemicals such as insecticides and anti-nerve gas resulted in gut microbial 78 dysbiosis. There has been a decrease in the relative abundance of several beneficial 79 bacteria. Recently, a study in GW-CMI veterans also reported similar alteration of gut 80 microbiome [24], [25], [26]. With reported studies on microbial dysbiosis patterns in the 81 preclinical animal models as well as the GW-CMI veteran well established, , we 82 hypothesized that exposure to environmental pesticides and pyridostigmine bromide 83 may also lead to an alteration of ARGs and MGE expression, an important constituent of 84 the gut resistome. In the present study we used both GWI veteran stool samples and 85 fecal pellets from a GWI persistence mouse model mimicking the present day veteran 86 health to study the alteration in gut resistome. We also aimed to study the possible 87 associations between gut resistome and CMI proinflammatory pathology and 88 mechanistically linked the altered resistome with a proinflammatory phenotype by using 89 a fecal microbiota transfer in mouse models.  Carolina at Columbia, SC. All the mice had ad libitum access to food and water and were 106 housed at 22-24°C with 12h light/12h dark cycles. The mice were sacrificed after the 107 animal experiments. Organs including frontal cortex and distal part of small intestine were 108 collected after dissecting the mice and fixed in Bouin's solution and 10% neutral buffered 109 formaldehyde respectively. Serum was collected from fresh blood of mice by performing 110 cardiac puncture after anesthesia. The fecal pellets were collected from colon and it was 111 stored at -80°C for whole-genome sequencing.

112
Mouse model of CMI related to Gulf War exposures. 113 After one week of acclimatization, the mice were randomly distributed into three groups.

114
The first group received vehicle (0.6% dimethyl sulfoxide) for two weeks and were 115 denoted Control (n=6). The second and third mice groups denoted GWI (n=6)  As part of the call back study, the veterans also filled out questionaires about their current 153 and recent gut health and use of antibiotics or probiotics.

155
The GWIC participants of the microbiome study were mailed a Second Genome stool 156 collection kit (Second Genome, San Francisco, CA, USA). The kit was a self-collecting 157 kit which contained a bar coded vial with stabilizing solution for long term preservation of 158 nucleic acids in stool during transportation and storage. Once received from the subjects, 159 the stool samples were stored at -20°C and upon collection of significant sample 160 numbers, they were sent for whole genome shotgun sequencing by COSMOSID. The   [38] to produce three sets of bins. The "bin_refinement" module was used to refine these 186 three bin sets to produce a single set of best bins. Finally, the single bin set was used by 187 the "bin_reassembly" module which extracts the reads mapping to each bin and uses 188 them for a second round of de novo assembly to improve the completion and reduce the 189 contamination of the bins.   Analyses were performed using R v3.6.3 (1). AGF, MGE, and OTU count data were 235 normalized based on library size using the "estimateSizeFactors" function from the 236 "DESeq2" package [45] with the parameters "type = poscounts". Normalized count data 237 were then log transformed with the base R function "log2". PERMANOVA was calculated 238 using the "adonis" function from the package "vegan" (2) with Bray-Curtis dissimilarity and 239 9999 permutations. Welch two-sample t-test was implemented using the base R function 240 "t.test" with the parameters "conf.level = 0.95, alternative = two.sided" indicating 95% 241 confidence level and two-tailed testing. PCA ordinations of AGF and Taxonomy 242 abundance data were performed using the "rda" function from "vegan". Procrustes 243 analysis was performed using PCA ordinations with the function "protest" from "vegan" 244 with 9999 permutations. Distance-based redundancy analysis was performed with the 245 "capscale" function from the "vegan" package with Bray-Curtis dissimilarity. Unpaired t-246 tests (two-tailed tests with equal variance) were performed followed by Bonferroni-Dunn 247 post-hoc analysis to compare between the mouse experimental groups and veteran 248 groups respectively. Chao1 α-diversity was calculated using the "chao1" function from the 249 "fossil" package. Correlation analyses between α-diversity and selected biomarkers were 250 performed using Pearson's correlation implemented by the base R function "cor". All 251 visualizations were rendered using the "ggplot2" package unless otherwise described.

252
For all analyses, p ≤ 0.05 was considered statistically significant and data are represented 253 as mean ± standard error of mean.

255
Characterization of Anti-microbial resistance in mouse fecal samples 256 We performed whole genome shotgun sequencing, metagenomic assembly, and 257 functional gene annotation on fecal samples collected from 3 groups: Control, GWI, and 258 GWI_FMT) to construct AGF profiles associated with the GWI mouse model and the fecal 259 microbiota transplant FMT treatment. We detected 95 unique AGFs with an even 260 distribution across all groups with 185 ± 4 (mean ± standard error) total AGFs in Control, 261 182 ± 3 total AGFs in GWI, and 188 ± 4 total AGFs in GWI_FMT (Fig. 1a). We used 262 permutational multivariate analysis of variance (PERMANOVA) to compare the AGF 263 profiles between sample groups which revealed a significant deviation between the 264 GWI_FMT and GWI (R^2 = 42.2%, p = 0.002) and GWI_FMT and Control groups (R^2 = 265 41.3%, p = 0.0001). This analysis indicates that the sample group variable explains 266 roughly 40% of the variation in the resistome profile. To get a clearer picture of the specific 267 changes in the resistome, we performed differential abundance analysis of the AGFs by 268 manually selecting the AGFs with the greatest variance in relative abundance between 269 sample groups and compared their sum relative abundance as a subset of the resistome 270 (Fig. 1b). These selected AGF groups were significantly increased in the GWI group when between the Control and GWI groups in the CAP2 axis suggesting that FMT treatment 279 may act to return the resistome to a more Control-like state (Fig. 1D).

280
To assess the transferability of the AGFs in the microbiome, we also examined the profile 281 of the MGEs constituting the mobilome in each sample group. We detected 67 unique 282 MGEs evenly distributed across all groups with 74 ± 5 total MGEs in Control, 71 ± 3 total 283 MGEs in GWI, and 80 ± 3 total MGEs in GWI_FMT (Fig. 1e). PERMANOVA showed that 284 significant differences in the mobilomes of the GWI_FMT and GWI groups (R^2 = 42.9%, 285 p = 0.0022) and the GWI_FMT and Control groups (R^2 = 30.6%, p = 0.0024). The 286 relative abundance of manually selected MGEs were also significantly different between 287 Control and GWI groups (Welch two-sample t-test, p = 0.002101) and between the GWI 288 and GWI_FMT groups (Welch two-sample t-test, p = 0.0003543).

Distribution of ARGs and MGEs across different mouse groups 290
In our analysis of the gut resistome, we detected multiple genes AFGs imparting 291 resistance to antimicrobial classes which have been, marked as highly important and 292 critically important by the World Health Organization WHO (AGISAR, 2018). Overall, 293 glycopepetide resistance genes were the most common type of AGF followed by 294 quinolones, polymyxins and carbapenem resistance genes. The relative abundance of 295 vancomycin resistance genes were highest (4.44% van-R, 3.57% van-S, 2.33% van-Y 296 and 1.9% van-T) in GWI group compared to Control and GWI_FMT groups (Fig 2a).

297
Interestingly, there were 38 unique genes which were only present in the GWI group but 298 not in the Control or GWI_FMT groups. Of these genes, mcr-1 and ndm-1 drew the most 299 concern because of their high transferability. The Control group also showed maximum 300 resistance towards vancomycin followed by RND (resistance nodulation cell division) 301 efflux pump, MATE transporter and β-lactamase resistance genes. Apart from 302 vancomycin-methicillin resistance, resistance against penicillin, cephalosporins, and 303 macrolides were also observed in GWI_FMT samples. Though these mice were not 304 exposed to any of these antibiotics, and the very fact that they showed resistance towards 305 members of critically important the highest priority group of antimicrobials (AGISAR, 306 2018,WHO) is a matter of concern. 307 When comparing the different drug class resistance across the three groups, the majority 308 of the resistome was composed of genes imparting mixed resistance against drug classes 309 with the largest single class being glycopeptide resistance. Genes against this drug class 310 consisted of 4.6% and, 4.4%, and 3.83% of the GWI, Control, and GWI_FMT resistome 311 respectively. Apart from the observed glycopeptide resistance, other similar fractions of 312 the resistomes detected were in relation to macrolide antibiotics with 4.14%, 4.08%, and 313 3.74% respectively, peptide antibiotics with 4.06%, 3.94%, and 3.45% respectively, 314 tetracycline antimicrobials 4.08%, 4.01%, and 3.75% respectively and nitroimidazole 315 antibiotics which were present 3.76%, 3.63%, and 3.12% respectively in the GWI, control 316 and GWI_FMT groups (Fig 2b). Multivariate analysis of the profiles of drug class 317 resistances showed significant deviation and correlation dependent on sample groups 318 when comparing the GWI_FMT and Control groups (PERMANOVA, R^2 = 40.9%, p = 319 0.0027) and the GWI_FMT and GWI groups (PERMANOVA, R^2 = 56.4%, p = 0.0025).

320
Typical gene signatures are identifiable for classes of resistance mechanisms. 321 Resistance mechanism genes such as those which are known to be associated with 322 antibiotic efflux were identified and and made up the highest percentage of all resistomes 323 (20.33%, 20.15%, 19.75% in GWI, control and GWI_FMT respectively) (Fig 2c). Other 11.8% in GWI_FMT). Tn 916 (9.3% in GWI), integrase (11.2% in GWI), ISCR (9.57%) 335 were also present in all the 3 groups of mice (Fig 2e). Multivariate analysis of the MGE 336 type profiles showed a significant deviation and correlation pattern that was dependent 337 on sample groups when comparing to the GWI_FMT and Control groups (PERMANOVA, 338 R^2 = 42.2%, p = 0.0024) and the GWI_FMT and GWI groups (PERMANOVA, R^2 = 339 52.2%, p = 0.0018).

341
In order to study whether the results observed on our mouse model were mirrored in the 342 human subjects, we obtained 15 stool samples from GWI veterans (5 in the Hum_Control 343 group and 10 the in Hum_GWI group). We then performed whole genome shotgun 344 sequencing, metagenomic assembly, and functional gene annotation. We detected 108 345 unique AGFs with 175 ± 3 and 182 ± 3 total occurrences in the Control group and GWI 346 groups respectively (Fig 3a). On comparing the relative abundance of AGFs, we observed 347 a modest but insignificant deviation in the resistome profiles between the two groups 348 (PERMANOVA, R^2 = 7.2%, p = 0.4428). When looking at a more granular scale, there 349 was an insignificant decrease in the relative abundance of selected AGFs (Welch two-350 sample t-test, p = 0.6874) (Fig 3b). Procustes analysis showed a significant correlation of 351 bacterial taxa and the AGFs (PROTEST, M^2 = 0.5972, p = 0.004) (Fig 3c). Despite minor 352 divergence indicated by multivariate analysis, distance-based redundancy analysis 353 showed clear clustering of the control and GWI groups with minimal overlap on the first 354 two constrained principal coordinates (Fig 3d).

355
Looking at transferability, we identified 92 unique MGEs across both groups with 77 ± 5 356 and 85 ± 3 total MGEs in the Control group and the GWI groups respectively (Fig 3e).

357
While there was some deviation in the MGE profiles of the two groups, multivariate 358 analysis revealed that these changes were not significantly dependent on the sample 359 groups (PERMANOVA, R^2 = 6.5%, p = 0.599). Interestingly, the relative abundance of 360 selected MGEs were higher in GWI group compared to the Control, however, this 361 difference was not statistically significant (Welch two-sample t-test, p = 0.05331) (Fig 3f).

363
To check the distribution of ARGs and MGEs in human samples, identical analysis was 364 carried out in the 2 groups of human data. Ciprofloxacin antibiotic efflux pump showed 365 the highest relative abundance in GWI (5.26%) followed by RND antibiotic efflux pump 366 (4.16% in GWI), vanR (3.96% in GWI) and tetR (3.65% in GWI) (Fig 4a).

367
Comparing the drug class resistances of the 2 groups, glycopeptide antibiotic was most 368 predominant followed by macrolide, tetracycline, fluroquinolone and glycycline antibiotics.

369
The percentage of the resistance were observed in the sequence of 4.08%, 4.01%, 3.91% 370 and 3.1% respectively. The profiles of drug class resistances did not significantly deviate 371 or depend on the sample group (PERMANOVA, R^2 = 6.4%, p = 0.6181) (Fig 4b)

372
A total 9 different types of resistance mechanisms were detected, in which antibiotic 373 efflux pump (20.72 % in GWI and 20.76% in control) was the most common. The 374 mechanisms of antibiotic target alteration, protection, inactivation and replacement 375 mechanisms were also detected across the 2 groups (Fig 4c).

376
Among the 31 unique different MGE groups studied, tnp A was found to be most 377 predominant (7.8% in GWI,8.8% in Control). A total of 23 MGEs were found to be present 378 in the GWI group. Int2, IS 91, IS 621 and MGE 10 were found to be present in both 379 groups (Fig 4d). We also observed 8 different MGE types unique in GWI group but transposases (13.63% in GWI and 12.71% in control) and integrases (10.67% in GWI 381 and 9.97% in control) were found to be the most abundant in all MGE types (Fig 4e).

382
Multivariate analysis of the MGE class profiles revealed moderate deviation between 383 sample groups which was shown to be independent when compared between sample 384 groups (PERMANOVA, R^2 = 12.1%, p = 0.0632).

385
Expression study of antimicrobial resistance genes in mouse and GWI veteran samples LRA β-lactamase (LRA) were increased by 3.1 and 2.15 fold when compared to FMT groups suggesting that fecal microbiota transfer was efficient in decreasing the 403 abundance and expression of these antimicrobial resistance genes (Fig 5a & b).

404
Comparison of microbiome sequencing data in human samples showed that there was Control group, as shown by immunoreactivity of the cytokine in the villi (Fig 6a,b). Treatment with FMT significantly decreased (p<0.000001) the expression of IL-1β in 426 GWI_FMT mice groups compared to GW group (Fig 6a,B). To study the association 427 between IL-1β expression and AGF diversity, we performed a correlation analysis.

428
Results showed a positive correlation (r= 0.8463, p=0.034 & r=0.9691, p=0.001 429 respectively) between α-diversity of AGFs, resistant drug classes and IL-1β, suggesting 430 that alteration of gut resistome had a significant correlation in gastrointestinal 431 inflammation (Fig 6f). 432 We observed a significant increase (p=0.000413) in serum IL-6 level in GWI mouse group 433 when compared to Control and GWI_FMT group (Fig 6c). Results also suggested that  (Fig 6g).

437
Our previous studies have reported that a decrease in synaptic plasticity marker BDNF 438 played a key role in brain pathology in GW chemical exposed mice [44] Results showed 439 that expression of BDNF significantly decreased (p=<0.00001 between Control vs GWI & 440 GWI vs GWI_FMT) in GWI group when compared to Control and GW_FMT mice groups 441 (Fig 6d). Interestingly, a negative correlation (r=-0.8866, p=0.19 & r=-0.9799, p<0.001) 442 was observed between BDNF and AGFs suggesting that increased AGF-α-diversity may 443 have a strong influence on observed neuroinflammation GWI mice group (Fig 6j)  the present study, we treated GW chemical exposed mice with FMT in GWI_FMT group. 500 We observed a significant decrease in relative abundance of selected AGFs and MGEs are present in GWI mouse group but not in Control and GWI_ FMT group. As mentioned 510 previously, these mice were never exposed to any kind of antibiotics (β-lactams or 511 colistin) during the entire course of the study, but they showed a spontaneous acquisition 512 of AGFs in GW chemical exposed group. This may be due to intrinsic transfer or extrinsic 513 transfer through HGT. We were able to confirm the presence of these genes by 514 sequencing (data not shown), but further studies are needed to study their mechanism of 515 transfer. Another limitation in our observation of gene transfer may be from the use of diet 516 in these mice. The mice in our study were fed with the standard chow diet though in a 517 consistent manner in all groups. Diet induced acquisition of microbial resistance genes 518 should be studied as a mechanism of acquiring antibiotic resistance genes in all further 519 studies though the chemical exposure was the principal variable that differentiated the 520 groups.

521
Here in this study, we were able to identify diverse types of MGEs in which tnpA is most 522 abundant followed by transposase. Another study by Parnanen et

543
In conclusion, our study revealed for the first time that exposure to GW chemicals 544 associated with CMI caused significant changes in ARGs and MGEs in GW-CMI mouse 545 model and veterans alike. Further , a strong association was stablished between an 546 altered resistome and systemic IL6 levels in a translatable mouse model that has broad 547 implications in the general population suffering from similar ailments. Strikingly it is 548 expected that a strong 78 million of the US population will be elderly category by 2030.

549
Most of them have a history of prolonged antibiotic use, a case similar with our aging 550 veterans. The scenario is also striking owing to old age associated hospitalizations and 551 increased chances of hospital acquired infections. In addition, FMT can be used as a 552 therapeutic strategy against the increased antibiotic resistance in veterans and elderly to 553 attenuate a possible altered resistome. A focused study on each of the antibiotic resistant 554 genes that was altered in the CMI mouse model and its modulation by FMT or in GW-555 CMI veterans can be an avenue for drug discovery and personalized medicine in future.